qsimd_p.h doesn't like when you compile with just -mavx2. I guess no one
is using these flags except me, and I already set the full arch in
CFLAGS.
Qt 6.4 calls x86-64-v3 "arch haswell". Will be fixed in 6.5.
Drive-by remove the AVX512 bits that weren't in use.
Pick-to: 6.4
Change-Id: I76216ced393445a4ae2dfffd172a4f5684351514
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Mike Krus <mike.krus@kdab.com>
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
FS's input qualifier is different with VS's in GLSL1.10.
If prototypes has 'headerSnippetsFrag', it will be applied for FS
instead of 'headerSnippets'.
Fixes: QTBUG-99019
Pick-to: 6.5 6.4 6.2
Change-Id: I588d3f78d4ba895a9a90b20a87b92a6da9ffee18
Reviewed-by: Janne Koskinen <janne.p.koskinen@qt.io>
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
The STL containers used in some types return size_t as their size(),
which gets signed/unsigned comparison warnings. Use unsigned literals
for the constants to compare against.
Change-Id: I91887a1dbd3eb95655c86add2e9f589f0c970def
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
loop variable has narrower type 'ushort' (aka 'unsigned short') than
iteration's upper bound 'unsigned int' [bugprone-too-small-loop-
variable]
Change-Id: Id6ab16316f9eb40e47d8494bf70621792284a7b4
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
openglvertexarrayobject.cpp:117:5: this call will remove at most one
item even when multiple items should be removed [bugprone-inaccurate-
erase]
Change-Id: Ife88facc1cad159658c08180548fe365eaeb1714
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
Change the variable type from int to qsizetype when receiving qsizetype
values.
Change-Id: I2adf71e1d2f26b1452fee28890c0f68300f54224
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
When using OnDemand, we detect whether Qt3D actually needs to render or
not. Based on that we set a flag on the Scene3DRenderer to tell it whether
it should render or not.
The subtlety lies in the fact that depending on that flag value, we will
still call Qt3D render but not at the same moment. We need to call render
to unlock the Qt3D render loop for next frame.
Therefore when we should skip a frame, we call render in
Scene3DRender::beforeSynchronize (updatePaintNode) and return early
while in the other case beforeSynchronize doesn't call render and it gets
called later in beforeRenderPassRecording. We know that the render call
in the skip frame case won't render anything.
To work correctly, we must obviously set the skip frame flag before
beforeSynchronize / beforeRenderPassRecording is called which
wasn't the case. This therefore resulted in random screen flashing as we
would end up having beforeSynchronize rely on the skip frame value for the
previous frame. This means we would end up issuing a render call in
beforeSynchronize using assuming it wouldn't render anything when it actually
did (no FBO would be bound and Qt3D would render straight to the window
...)
Also update the number of frames to render before skipping to 3 to match
Qt 5.15 and ensure we have something visible for the first time
Scene3D is launched.
Pick-to: 6.4 6.5
Change-Id: Icc34b62a0af2ff44232b39c694a4702488a890a2
Reviewed-by: Mike Krus <mike.krus@kdab.com>
There are 2 ways to dispose of a QRhiResourceUpdateBatch obtained with
QRhi::nextResourceUpdateBatch():
1) Either give it to beginPass/endPass/beginComputePass/endComputePass
so it is sent to the RHI backend and then automatically released;
2) or call QRhiResourceUpdateBatch::release() if the resource update
is not used for any pass.
The Qt3D RHI renderer was never disposing of the
QRhiResourceUpdateBatch, and could end up exhausting all the 64 batches
available in the RHI: "Resource update batch pool exhausted (max is
64)", in addition to consuming a lot of RAM if the buffers or textures
are large.
To fix this, ensure the QRhiResourceUpdateBatch'es we allocate are always
released: when requesting a new one, send the previous one to
endPass/endComputePass (1), and when destroying
the SubmissionContext, release the one we will not use (2).
Pick-to: 6.4 6.5
Change-Id: Ic3fb145b11a8e615c83745f560f019e6c4ee37f7
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
In several places, std::make_move_iterator was used to move from a const
vector, which is not possible and fell back to copy.
Use the existing Qt3DCore::moveAtEnd utility to ensure a move operation
is actually happening.
* When readability permits, pass directly a return value as second
parameter;
* Otherwise, std::move from a mutable vector.
Change-Id: I56c8925bbdeab95257d0a7fa89cffcda53fd6451
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
- Disable the examples that use QtMultimedia as they haven't been ported
to QtMultimedia from Qt6
- QTextureDataUpdate forward header doesn't seem to be generated
- Add QuickWidgets in the find_package call so that we can later check
if it is actually available
Pick-to: 6.4 6.5
Change-Id: I98d000e839dd0cb0aad862b66bfe8b9725a13744
Reviewed-by: Mike Krus <mike.krus@kdab.com>
syncqt doesn't find the class name if the namespace line doesn't end
with {.
Pick-to: 6.4 6.5
Change-Id: Icf30464ed0ecb9ce84d6c62c04edfc5d374ed26f
Reviewed-by: Mike Krus <mike.krus@kdab.com>
There is a risk that the front-end QSceneLoader node has been destroyed
by the time the LoadSceneJob completes. The postFrame() would then do
nothing, and the m_sceneSubtree pointer will be forgotten, leaking the
whole tree. This can happen in real life when loading very large scenes.
To ensure that we cannot leak the scene tree, use strict pointer
ownership with std::unique_ptr.
Pick-to: 6.4 6.5
Change-Id: Ie2281bc178fc8793bab967a13ea8d30aa46268a0
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
Sometimes the shader manager returns a nullptr when fetching the shader from a handle, so the return value needs to be checked before it is accessed.
Pick-to: 6.5 6.4 5.15
Change-Id: Ia021cd8f22ceb2626a7a2734b7e346fbcc8e0301
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
The Shaders or the QShaderProgramBuilders may not exist in the managers
anymore by the time we arrive in
Renderer::sendShaderChangesToFrontend(). This happens when Qt3D shuts
down.
To fix this, check the returned values by the managers.
This is similar to the existing patch for the OpenGL renderer:
Ia021cd8f22ceb2626a7a2734b7e346fbcc8e0301
Pick-to: 6.4
Change-Id: I322bf2524016c1f98433891ad230b65272fabbdd
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
The Renderer seems to already take care of deleting the abandoned
shaders thanks to the APIShaderManager::purge() method. However, when
Qt3D is shut down, there remain active RHIShaders in the
RHIShaderManager::m_apiShaders, that nobody deletes.
To fix this, in the RHIResourceManagers, delete the remaining RHIShaders
as we do with the other RHI resources.
Pick-to: 6.4
Change-Id: Iccc5470831eb7df0e120db6864fae425005c6919
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
In Scene3DItem::updatePaintNode(), in the Render thread, we try to set
the parent of the aspect engine being deleted to nullptr. This was
triggering an assert in QCoreApplication::sendEvent(), since the aspect
engine lives in a different thread, the Main thread.
Don't trigger this assert by avoiding to send ChildAdded/ChildRemoved
events when reparenting the aspect manager, thanks to the
QQml_setParent_noEvent utility.
Pick-to: 6.4
Change-Id: Ic816701ee65654f7b18b4998c54feb4840af0a14
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
The multithreaded job doesn't check the visited bounding volumes for
validity before adding them to the list to process further. This leads
to crashes, namely nullptr-dereferences.
Fixes: QTBUG-108405
Pick-to: 6.4
Change-Id: I16a7c008c7ac4f04be40bdbba05d8c745c1c825e
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
Sceneparser plugins are not loaded anymore until a QSceneLoader is added
to the tree, which improves Qt3D startup time. Note: the load is now
done in the QAspectEngine "Main" thread instead of the QRenderAspect
"Render" thread.
Fixes: QTBUG-59468
Change-Id: I45482ba06acf4dbdf7d4ce2b1a4e06d5a09c6de9
Reviewed-by: Paul Lemire <paul.lemire@kdab.com>
qtattributionsscanner expects file paths to be separated by a space.
Pick-to: 6.2 6.4
Change-Id: I983870ea403983f2fc32bf0d986e76235d74e497
Reviewed-by: Jörg Bornemann <joerg.bornemann@qt.io>
Fix "Assigning instead of Comparing" in
tst_QResourceManager::heavyDutyMultiThreadedAccess()
Task-number: QTBUG-107693
Pick-to: 6.4
Change-Id: I018bf83dc7a3cc9d6faa654a73b2d460c3e4498a
Reviewed-by: Mike Krus <mike.krus@kdab.com>
allocated in Renderer::setBindingAndShaderResourcesForCommand
Pick-to: 6.4
Task-number: QTBUG-106972
Change-Id: I4d0a64f1fbb0e9037b1e4c930f91e1b55412f180
Reviewed-by: Mike Krus <mike.krus@kdab.com>
Thiago Macieira
Qt Submodule Update Bot
Inho Lee
Edward Welbourne
Alexey Edelev
Aurélien Brooke
Ville Voutilainen
Paul Lemire
Martin Andersson
Jani Heikkinen
Marc Mutz
Christian Riggenbach
Mike Krus
Liang Qi
Kai Köhne
Aurélien Brooke
Tasuku Suzuki